Researchers evaluated two machine learning models, ArchesWeather and ArchesWeatherGen, for their ability to simulate climate over long periods. The models were originally designed for short-term weather forecasting but showed promise in longer-term simulations when provided with additional boundary conditions such as sea surface temperature and sea ice cover. The study compared the models' performance against numerical climate models and found that they could reproduce large-scale circulations, interannual variability, and even capture extreme events. However, it's worth noting that these results are based on a specific experimental setup defined by the AI Model Intercomparison Project (AIMIP) Phase 1 protocol. --- Why it matters: This study matters to researchers in AI because it demonstrates the potential of machine learning models for long-term climate simulation, which could have significant implications for climate modeling and prediction. The findings suggest that these models can be adapted for longer-term applications with some modifications. Source: https://arxiv.org/abs/2605.29976